OBJECTIVES: The objective of this study was to determine if routinely collected administrative claims data could be used to effectively predict what segment of a population of patients with type 2 Diabetes Mellitus (type 2 DM) would progress to insulin as part of their drug regimen. METHODS: To determine the time until a patient with type 2 DM starts insulin, defined as survival time, we used the PHREG procedure of SAS. This procedure uses Cox's proportional hazards model in order to estimate survival functions for diabetic patients. Based on the number of medications in the patients' regimen, eleven models were developed to predict the number of patients in a cohort expected to start insulin therapy over the two-year study period. The models were also used to identify the patients most likely to start insulin therapy and to estimate their probability. Split sample design was used to gauge the predictive ability of the models. RESULTS: In the monotherapy cohort model, the average of the absolute difference between the predicted and actual number of patients starting insulin each month was 0.965, with the maximum error for any month being 4.1 patients (an average of 27 patients started insulin per month). 27.03% and 24.12% of the patients that went on insulin within six months or two years respectively were in the top 10% in terms of risk. In comparison, 3.3% and 3.53% of the patients that went on insulin within six months or two years respectively were found to be in the bottom 10% in terms of risk. CONCLUSIONS: This study demonstrates that survival models can be used to predict and identify patients with type 2 DM who will require insulin as part of their treatment regimen. As a result, it is possible to develop tools based on these models that can be used by practitioners to assist in patient care.